Improving pitch detection through emphasized harmonics in time-domain

초록

In speech signal processing, it is crucial to detect the accurate pitch period of voice in time domain. The concept of pitch period is being utilized in various fields, including systems for speech enhancement, automatic speech recognition, speaker classification, and even voice guiding for the visually impaired. The periodicity of a voice signal has been emphasized in order to detect the pitch period more accurately as we can see in the current techniques, such as 'peak and valley technique,' 'auto-correlation method,' or 'center-clipping and signal-square.' However, all of these methods present a problem in finding accurate pitch period due to noise as well as the transitional section between voiced and unvoiced sounds. This paper proposes an improved method for detecting pitch period in time domain more accurately by using emphasized harmonics through non-linear clipping and synthesis technique. © 2012 Springer-Verlag.